机器学习驱动的农产品进销决策--以新疆棉为例

Machine Learning-driven Purchase-sale Decision-making for Agricultural Products--A Case of Xinjiang Cotton

我国是农产品生产和消费大国。近年来,基差交易成为大宗农产品的主要交易形式。由此,农产品经销企业面临采购和基于基差的销售定价决策。然而,不同于工业产品,农产品存在显著的规格不一致问题,给决策带来了困难。为此,首先,构建了基于基差的农产品进销两阶段非线性随机规划模型;其次,对模型进行线性化处理,并设计L-shaped算法以求解模型的最优解;然后,利用历史期货价格和交易状况大数据沉淀,采用多种机器学习方法预测模型所需的随机场景,实现了“从预测到决策”的研究;最后,以新疆棉为例,基于某经销企业实际数据开展实例应用与分析。本研究有助于提升大宗农产品交易效率和效益,并为包括新疆棉的我国棉花产销决策提供了科学支持。

China is a substantial producer and consumer of agricultural products.In recent years,basis trading has become one of the main trading ways of bulk agricultural products.Agricultural distributors face the decision-making on purchase quantity and sale pricing based on basis trading.However,different from industrial products,there is a significant issue of inconsistent specifications in agricultural products,which brings difficulties to decision-making.Firstly,a two-stage non-linear stochastic programming model was developed for the purchase-sales of agricultural products based on basis.Secondly,the model was linearized,and an L-shaped algorithm was designed to give the optimal solution.Thirdly,based on big data of historic futures prices and trading information,multiple machine learning approaches were employed to forecast scenarios that the model required.Thus,the study“from predictive to perspective”was implemented.Finally,the proposed approach was applied to address the real-world case of a trading company for Xinjiang cotton.The results can help to improve the effectiveness and profit of bulk agricultural product trading and provide scientific support to the decision-making of cotton trading in our country,including Xinjiang cotton.